2 October 2026
v1.10.0Sharing a conversation
Send the link to a conversation to other users on the platform so they can read it or carry it on. With view access, they see the conversation as it stood when you shared it. With edit access, they continue on a branch of their own, so your original conversation is left untouched.Publishing an artifact as a public page
Turn what the agent produced into a web page with its own link, to send to customers or anyone without an account, no sign-in needed. You can set a password or a time limit on access, and unpublish whenever you want.Highlight to ask a follow-up
Highlight any passage in an agent answer, choose Ask and type your question. The selected passage is attached to the question, so the agent answers about exactly that part while keeping the thread of the conversation. Copying text works just as before.Turning an agent’s default capabilities on or off
Keep or switch off each of an agent’s default capabilities, per version, so the agent does only the work you need from it.Tidier multi-step work
With models that reason, the agent now carries its line of thought from one tool step to the next instead of planning again from scratch after each step. It goes round in circles less, finishes sooner and uses fewer resources. Other improvements- Live Chat answers display formatted text, files, links, images and carousels directly.
- Agents can answer from knowledge published to a company or department when a member of that unit is chatting.
- Documents uploaded to knowledge from Console go through the standard document processing pipeline, for steadier search results.
- Knowledge cards on Console show the exact date and time of the last update.
- The icons and names of the tools the agent uses while working are now correct and consistent.
- Guardrails also check the agent’s answers, blocking content that breaks the policy even when a question tries to slip around it.
30 September 2026
v1.9.2Scheduled runs for agents
Set an agent to handle recurring work on its own: Create a task with the request, pick a daily or weekly cycle and the time it runs. When the time comes, the agent does the work and posts the result into your conversation. You can see the history of every run, and pause or resume it whenever you want.Charts right inside the conversation
When an answer carries data, the agent draws the chart in the chat instead of just listing numbers. Bar, line, pie, area and scatter charts are supported, each with a data table alongside for checking. Hover a point to read the exact value; charts follow the light or dark theme.Suggested follow-up questions after every answer
Under each answer there are a few suggested questions that follow what you are discussing. One click sends it, so you do not have to work out what to ask next. You can turn the suggestions off in settings.Processing steps collapsed into one group
The narration and the tools an agent calls while working are now gathered into a single collapsed group that you open when you want the detail. Each step is also named in plainer language.Messages sent back to back keep their order
When a customer sends several messages in a row on an external channel while the agent is still replying, the messages are handled in order instead of overlapping. An image and the text split across two messages are also merged into one request.MCP connections with OAuth 2.1
MCP servers now connect over the OAuth 2.1 standard, with no dependence on any single provider. Other improvements- Chat shows the context compaction status with a progress bar and a line recorded in the conversation history.
- Refine with AI picks up configuration files that were edited by hand between turns.
- Reloading the page restores every file the agent created in the chat.
- The chat no longer jumps while reloading: Image frames keep their size and the content is laid out in advance.
- Confirmation questions no longer disappear with an error when you have not answered yet.
- Updated cover images and fixed the getting started cards on the Console home screen.
- The description of the allowed domain list for embedding the chat widget now matches the actual behaviour.
- Files waiting to be deleted no longer appear in knowledge search results.
24 September 2026
v1.8.0Web search and Deep Research mode
In the chat on Workspace you turn web search on or off for each conversation; when a question needs information from the web and the feature is off, the agent reminds you to turn it back on. Deep Research mode lets the agent consult several sources at once and come back with a consolidated report instead of answering straight away.Conversation history stays with the agent
An agent shared across workspaces now keeps its conversation history on the agent itself, so you can review all of its conversations no matter which workspace you opened it from.A smoother Live Chat experience
Agent answers now appear smoothly, word by word, instead of jumping in chunks. The cursor stays in the input box after each send, so a customer can write the next message and attach files while the agent is still replying. The chat widget also runs only on the domains you have declared, which keeps control over where it is embedded. Other improvements- Uploaded skill packages keep every file type, and the agent can read binary files that come with the package.
- Fixed the Guardrail “View details” window showing the wrong topic and description.
- Fixed links and cover images in the getting started section on the Console home screen.
22 September 2026
v1.7.0A personalised experience with your agent
Tell the agent your nickname, your job and anything else you want it to know, so its answers fit what you need. The agent remembers things from your conversations on its own; you can ask it to update them, edit them directly, delete them, or turn memory off at any time. Memory is stored separately for each workspace.Branch into a new conversation
From any answer in the history you can split off a new conversation to try a different direction without touching the original thread. The new branch keeps the context and attachments up to the turn you branched from, and deleting one side does not affect the other.Deploy an agent to Microsoft Teams
Agents talk to customers directly on Microsoft Teams, with a configuration screen per channel on Console. Teams also gains filters and message indicators that make conversations easier to follow.Connect third-party agents to Console
Connect agents built outside the platform into Console and use the same chat interface on Workspace. The connection configuration is versioned, with draft, publish and roll back to an earlier version; each version keeps its own credentials and the connection can be tested before you publish.Managing roles and members in an organisation
See the default roles with the permissions attached to each, invite members into the organisation and assign roles, then update or remove a role at any time. Permissions cover Connectors, Guardrails, Skills, Agents and roles. More in Organisation management.Choose the approval mode when refining an agent with AI
When you use AI to refine an agent you choose whether it applies its proposals automatically or you approve each step. Automatic is the default, and the system remembers your last choice.Send JSON and HTML in a conversation
The input box takes JSON and HTML directly, keeping the structure both on screen and when it is sent to the agent. Other improvements- New models added.
- Regenerating an answer and editing a question now replace in place on the last turn, with a switch to move between the versions you have generated.
- Download an agent answer as PDF, Markdown or Word.
- Conversation history also shows the tools the agent used, and a conversation can be found from part of its identifier.
- Added greeting configuration for Live Chat.
- Fixed the wrong date and time display and the preview that would not scroll when customising the Live Chat widget.
- Save confirmations no longer stay stuck on screen covering other content.
- Input suggestions in widget customisation now show in the default language you selected.
- Reviewed all Indonesian content across the agent.
- Web search and page loading are now two separate tools, which gives tighter, more accurate results.
18 September 2026
v1.6.0Declining an invitation
People invited to an organisation can now decline rather than leave the invitation hanging. Declined shows as its own column in the invitation list, so administrators can see at a glance who has joined, who has not replied and who has declined. More in Organisation management.Create a project with knowledge and conversations attached
When creating a new project you can bring knowledge documents and conversations in with it, instead of adding each piece afterwards.Default roles and permissions for new organisations
A newly created organisation arrives with a default set of roles and permissions, so an administrator can invite people and assign access straight away. More in Organisation management.Clearer test chat before release
The test chat panel in Console now shows the agent name at the top, in full. Each test chat is also saved as its own conversation in the agent’s history rather than being lumped together, so you can look back at individual test runs. Other improvements- Built-in connector actions keep their original tool names, which helps the agent call the right one.
- Removed redundant interface labels from the chat widget configuration.
16 September 2026
v1.5.0Voice input
Speak instead of typing: tap the microphone in the Workspace chat box and what you say is transcribed straight into the input for the agent.Deploying an agent to Live Chat
Live Chat is a new deployment channel: a ready-made chat window whose colours, greeting and suggested questions you set on Console. Take the agent to your customers by pasting a snippet into your website, sharing a link or QR code, or embedding it in an Android or iOS app. See Live Chat channel.Integrating an agent over the API
Already have your own chat interface, a self-run Zalo OA or a contact centre? The API channel lets your systems pass customer messages to the agent over a webhook and receive answers back over a callback, with support for buttons, quick replies, carousels and source citations. Every packet is signed. See API integration.Pausing a sub-agent
Disconnect a sub-agent without deleting it, and reconnect whenever you need it again. More in Sub-agents. Other improvements- Fixed a connection problem with the Zalo channel.
- The default content guardrail recognises unsuitable content more accurately and is harder to get around.
- Hardened security across all input.
- Agents can read larger attachments.
- Built-in agent templates display in your language and are more compact.
- Integration guides state which version they apply to.
11 September 2026
v1.4.0Deploying an agent on Zalo
Agents talk to customers directly on Zalo, with messages formatted to suit the channel. The agent can read files customers send over Zalo and supports a range of file types. See how to manage channels in Publishing.Managing companies, departments and guests
Organisation administrators create and manage companies, departments and the members of each. People outside the organisation can be invited into a company as guests, with nothing more than an email address. More in Organisation management.Default content guardrail
Agents created or edited in Console and Workspace now come with the system’s default content guardrail applied. You can still add your own rules per agent. More in Guardrails. Other improvements- Agents no longer hang for long stretches when a model stops responding mid-answer.
- Agents still return source citations when a connector cannot connect.
- Creating a new agent version no longer puts it into production automatically. A version only goes live when you publish it deliberately.
- Clearer error reporting when the agent’s code execution environment runs into trouble.
- Security hardening and updated interface wording.
8 September 2026
v1.3.0Suggested questions for agents
Set up to 3 suggested questions per agent in FPT AI Console. When a user opens a new conversation the questions appear under the input box - one click and they are away, with nothing to think up first. Other improvements- Security hardening across Workspace and FPT AI Console.
4 September 2026
v1.2.0Publishing agents by org / unit
Administrators grant access per org or unit, place agents into the right unit themselves, and approve before an agent is published to that unit. See Publishing. Other improvements- Agents can answer at greater length without being cut off mid-sentence.
- Code execution is faster and more stable.
- Clearer status when a model runs into trouble.
- Fixed connector status display in Console and a connector configuration bug for sub-agents.